Vibrations Summary

1. Free Vibrations
1.1 Introduction to Vibrations
Vibrations are often unwanted phenomena in aerospace engineering. When systems start vibrating at the
wrong frequencies, they might fail, which isnt particulary good. In reality all systems are continuous
systems, meaning that the displacements of parts depend on a lot of factors. To simplify this, the system
is often modeled as a discrete system. Here the system is split up in parts, which are then evaluated
separately.
Two types of vibrations can be distinguished, being free vibrations and forced vibrations. In free vi-
brations no energy is exchanged with the environment, while in forced vibrations there is energy exchange.
First we will have a look at free vibrations. Forced vibrations will be treated in later chapters.
1.2 Stiness of an Axially Loaded Rod
Lets consider an axially loaded rod of negligible mass, having a mass attached to its end. We know that
the displacement of the mass is given by
=
FL
EA
, (1.2.1)
where F is the (tensional) force in the bar, L is the length of the bar, E is the E-modulus and A is the
cross-sectional area. The stiness k is dened as the force needed to reach unit displacement. In an
equation this is
k =
F

. (1.2.2)
So for our axially loaded rod, we will have
k =
EA
L
. (1.2.3)
We can now model the situation. We do this by replacing the bar by a spring with the stiness k. This
is shown in gure 1.1.
Figure 1.1: Modeling of an axially loaded rod.
1
1.3 Motion of an Axially Loaded Rod
Previously we considered the axially loaded rod and modeled it. Lets turn to gure 1.1 once more. We
would like to know how the system will move, if it is given a certain initial displacement/velocity.
To nd this out, we use Newtons second law F = ma. The only force acting on the mass is the spring force
F
s
(we dont consider gravity yet). We know that the spring force varies linearly with the displacement
x by the stiness k. However, if the block moves upward, the spring forces points downward. So there is
a negative relation between the two. In an equation this becomes
F
s
= kx. (1.3.1)
If we combine this with Newtons second law, we will nd that
m x = F
s
= kx m x +kx = 0. (1.3.2)
The solution can be found by solving this dierential equation. We will get
x(t) = c
1
cos

k
m
t

+c
2
sin

k
m
t

. (1.3.3)
So the system will start vibrating with a xed angular frequency. This frequency, called the angular
eigenfrequency, is denoted by

n
=

k
m
. (1.3.4)
From this, the eigenfrequency f and vibration period T can be derived, according to
f =

n
2
=
1
2

k
m
and T =
1
f
=
2

n
= 2

m
k
. (1.3.5)
However, equation 1.3.3 isnt very useful. Instead, it is more meaningful to use
x(t) = Asin (
n
t +) , (1.3.6)
where A is the amplitude (usually taken to be positive) and is the phase. Both follow from the boundary
conditions. If we give the mass an initial displacement x
0
and an initial velocity v
0
, then we can nd A
and . They will be
A =

x
2
0
+

v
0

2
and = tan
1

n
x
0
v
0

. (1.3.7)
1.4 Eects of Gravity
Previously we havent considered gravity. What happens if we do? In this case the total force acting on
the mass will be F
s
+mg. This would turn the dierential equation into
m x +kx = mg. (1.4.1)
When solving dierential equations, we know we rst ought to nd the homogeneous solution of the
dierential equation
m x +kx = 0. (1.4.2)
We already know the solution for this. After we have found the homogeneous solution, we need to nd
one particular solution x
p
(t). Note that the non-homogeneous term mg is just a constant. So the
particular solution is probably constant too. It can then be shown that
x
p
(t) =
mg
k
. (1.4.3)
2
This makes the solution for the dierential equation
x(t) = x
h
(t) +x
p
(t) = Asin (
n
t +) +
mg
k
. (1.4.4)
Note that if the amplitude A is zero, then the mass will just have a constant displacement of mg/k. This
also follows from statics.
In vibrational engineering the homogeneous solution x
h
(t) is sometimes called the transient solution
x
tr
(t) and the particular solution x
p
(t) is also called the steady state solution x
ss
(t).
1.5 Motion of a Laterally Loaded Rod
Of course there are more kinds of vibrations then masses on axially loaded rods. Lets consider a laterally
loaded rod, as shown in gure 1.2. The rod has an (area) moment of inertia I.
Figure 1.2: Modeling of a laterally loaded rod.
This time the displacement , and thus also the stiness k and natural frequency
n
, are given by
=
FL
3
3EI
k =
F

=
3EI
L
3

n
=

k
m
=

3EI
mL
3
. (1.5.1)
The rest of the problem is similar to what we have previously discussed.
1.6 Rotation of a Torsionally Loaded Rod
Now lets consider an other case. We have a disk with (mass) moment of inertia J, connected to a rod
with (area) polar moment of inertia I
p
, as shown in gure 1.3.
We will be looking at the angular displacement . This depends on the moment M that is acting between
the rod and the disk. If this moment is known, then the angular displacement can be found using
=
ML
GI
p
. (1.6.1)
Now we can dene the torsional stiness as
k =
M

=
GI
p
L
. (1.6.2)
Note that the torsional stiness has as unit Nm, while the normal stiness has as unit N/m.
3
Figure 1.3: Modeling of a torsionally loaded rod.
Newtons second law for rotations states that M = J = J

. Combining this with the torsional stiness

gives us the dierential equation
J

+k = 0. (1.6.3)
We already know the solution to this! It is just
(t) =

sin (
n
t +) , (1.6.4)
where
n
=

k/J is the angular natural frequency and

denotes the amplitude of the vibration.
1.7 Other Cases
We have seen axially loaded rods, laterally loaded rods and torsionally loaded rods. There are, however,
innitely many other types of systems. It is, for example, possible to combine multiple springs in a
system. We wont be treating all those combinations, of course. If this is the case, the skills of the
engineer come into play.
However, were not letting you venture into those problems unguided. When face with a more complicated
system, just follow the following steps:
Consider the point of which you want to know the motion.
Express the force/moment at that point as a function of the (angular) displacement.
Use Newtons second law to nd the dierential equation.
Solve the dierential equation to nd the equation of motion.
1.8 Using Energy
In a free vibration (without damping), energy is conserved. You can consider two types of energy in a
vibration. These are kinetic energy T and potential energy U. Lets consider those energies for the
axially/laterally loaded rod. The kinetic energy is given by
T =
1
2
m x
2
. (1.8.1)
The potential energy here consists of spring energy and gravitational energy, and is given by
U =
1
2
kx
2
mgx, (1.8.2)
4
A very important rule is the rule of conservation of energy. It states that
T +U = constant = E, (1.8.3)
where E is the vibrational energy. If the mass passes through the equilibrium point, then T is maximal.
If the mass has maximum deection, then U is maximal.
It all sounds fun, but how can we use this? To use this, we dierentiate equation 1.8.3 with respect to
time. What we get is
dT
dt
+
dU
dt
= 0. (1.8.4)
If we work this out for an axially/laterally loaded rod, we will get
x(m x +kx mg) = 0. (1.8.5)
Note that x cant be zero for all t (or it would be an awfully boring problem). We now remain with
m x +kx = mg, (1.8.6)
which is exactly the dierential equation we needed to solve the problem.
The method that was just shown is called the energy method. When damping occurs, the energy
method is slightly more complicated. Now the lost energy also needs to be taken into account. We will
not treat this here though.
You may be wondering why we should use energy? Isnt it easier to just use Newtons second law? Well,
using Newtons second law is easier for normal one-dimensional problems. However, using energy when
solving multi-dimensional problems has various advantages. We will consider multi-dimensional problems
in more detail in the latest chapter of this summary.
5
2. Damped Motions
2.1 Introduction to Damping
The free vibrations discussed in the previous chapter dont stop oscillating. This isnt very realistic. So
we need to change our model. We therefore apply viscous damping. We assume that there is a force
acting on the mass in a direction opposite to the motion. This force is also proportional to the motion
(fast-moving objects have more friction). So we introduce the damping force
f
c
= c x(t), (2.1.1)
where the factor c > 0 is the damping coecient. If we combine this with the previous dierential
equation, we now get
m x +c x +kx = 0. (2.1.2)
To solve this dierential equation, we should rst solve the characteristic equation
m
2
+c +k = 0 =
c
2m

c
2
4km
2m
. (2.1.3)
The behaviour of the system now depends on the factor c
2
4km. Dierent things occur if this factor
is either smaller than zero, equal to zero or bigger than zero. Since this is so important, the critical
damping coecient c
cr
is dened such that
c
2
cr
4km = 0 c
cr
= 2

km = 2m
n
. (2.1.4)
Here
n
is the natural frequency of the undamped system, also called the undamped natural fre-
quency. We can now also dene the damping ratio as
=
c
c
cr
=
c
2

km
=
c
2m
n
. (2.1.5)
Note that since c > 0 also > 0. Using , the characteristic equation can be rewritten as
=
n

2
1. (2.1.6)
Three cases can now be distinguished, which will be treated in the coming paragraphs.
2.2 Underdamped Motion
In the underdamped motion the damping ratio is smaller than one. The solutions
1
and
2
of the
characteristic equation are now complex conjugates, being

1
=
n

1
2
i and
2
=
n
+
n

1
2
i. (2.2.1)
Before we write down the solution, we rst dene the damped natural frequency to be

d
=
n

1
2
. (2.2.2)
If we now solve the dierential equation, we will nd as the general solution
x(t) = Ae
nt
sin (
d
t +) , (2.2.3)
where A is the initial amplitude. Note that due to damping, the frequency of the vibration has changed.
The values of A and depend on the initial position x
0
and initial velocity v
0
and can be found using
A =

x
2
0
+

v
0
+
n
x
0

2
and = tan
1

x
0

d
v
0
+
n
x
0

. (2.2.4)
The underdamped motion results in an oscillation with a decreasing amplitude.
6
2.3 Overdamped Motion
In the overdamped motion the damping ratio is bigger than one. The roots to the characteristic
equation are now two real values, being

1
=
n

2
1 and
2
=
n
+
n

2
1. (2.3.1)
In this case no oscillation occurs. The mass will not even pass the equilibrium position. Instead, it will
only converge to it. Before we see how, we rst dene

c
=
n

2
1. (2.3.2)
The motion of the mass is now described by
x(t) = e
nt

a
1
e
ct
+a
2
e
ct

. (2.3.3)
The constants a
1
and a
2
once more depend on the initial conditions. They can be found using
a
1
=
1
2
x
0

1

n

v
0
2
c
and a
2
=
1
2
x
0

1 +

n

+
v
0
2
c
. (2.3.4)
2.4 Critically Damped Motion
In the critically damped motion we have = 1 and thus c = c
cr
. The roots of the characteristic
equation are now

1
=
2
=
n
. (2.4.1)
The solution is now given by
x(t) = (a
1
+a
2
t) e
nt
, (2.4.2)
where the constants a
1
and a
2
are given by
a
1
= x
0
and a
2
= v
0
+
n
x
0
. (2.4.3)
2.5 Stability
We have, up to now, considered only positive k and c. Of course it is also possible to have a negative
k (the force acts in the direction of the displacement or a negative c (the force acts in the direction of
motion.
If, for a certain motion, x , then the motion is unstable. Otherwise the motion is stable. We will
look at the stability of the systems for various combinations of c and k now.
k > 0 - This occurs in normal springs. In case of a deection, the mass is pulled back to the
equilibrium position.
For c = 0 we are on familiar grounds. The motion is just an undamped vibration. The
amplitude is bounded (x(t) A for all t) so we have a stable motion. However, x(t) never
converges to zero. So the system is only marginally stable.
For c > 0 we are dealing with a damped motion. It doesnt matter whether the system is
underdamped, overdamped or critically damped. In all cases x(t) 0 as t , so the
system is asymptotically stable. (How x goes to zero does depend on though, but this is
irrelevant for the stability.)
If c < 0, then the amplitude of the motion increases unbounded for increasing t. So the motion
is unstable. However, we can distinguish two cases.
7
If c
2
< 4mk (thus < 1), then there are still oscillations. In this case we have utter
instability.
For c
2
4mk (thus 1) no oscillation occurs. As soon as the mass departs from the
equilibrium, it will never return. Now there is divergent instability.
When k < 0 the mass gets pushed away from the equilibrium position, independent of the damping
coecient c. For c > 0 the motion only occurs slower than for c < 0. Since x(t) as t ,
the motion is unstable. To be more precise, there is divergent instability, since not a single
oscillation occurs.
2.6 Coulomb Friction
Suppose we have mass, horizontally sliding over a surface, as shown in gure 2.4.
Figure 2.4: Mass connected to a spring, sliding over a horizontal surface.
The force that acts on the mass depends on whether it is moving, and in which direction, according to
f
c
( x) =

N if x > 0
0 if x = 0
N if x < 0

= N

1 if x > 0
0 if x = 0
1 if x < 0

= Nsgn( x), (2.6.1)

where is the dynamic friction coecient and N is the normal force acting on the block. Also sgn()
is the signum function, dened to give 1 when > 0, 0 when = 0 and 1 if < 0. This kind of
damping is called Coulomb damping. The resulting dierential equation is
m x +Nsgn( x) +kx = 0. (2.6.2)
This is very hard to solve, due to the signum function. It is wiser to examine the problem in steps.
Suppose the mass has no initial velocity (v
0
= 0), but only an initial displacement
0
. If the initial
displacement is big enough to overcome the friction force (k
0
> N), the block will start sliding. After
/
n
seconds it will have reached a new maximum deection
1
. It can be shown that this deection is

1
=
0

2N
k
. (2.6.3)
If the force is big enough to let the block slide again, it will have another half oscillation of /
n
seconds,
but its maximum deection will have decreased again by the same amount. So,

2
=
1

2N
k
. (2.6.4)
This continues until after i half oscillations k
i
N. The block has been oscillating for i/
n
seconds.
But now the oscillation has ended and the block will remain at
i
.
8
3. Harmonic Excitation
3.1 Introduction to Harmonic Excitation
In the previous chapters, the only force present was the force of the spring. Although we also considered
gravity, this was a constant force and thus not very interesting. What will happen if we cause a time-
dependent external force F
e
(t) on the mass? In this case the dierential equation for an undamped
motion should be rewritten to
m x +kx = F
e
(t). (3.1.1)
We can get about any motion, depending on the external force. In reality external forces are often
harmonic. We therefore assume that
F
e
(t) =

F
e
cos t, (3.1.2)
where is the angular frequency of the external force. To solve this dierential equation, we rst
need to nd the homogeneous solution. This solution is already known from previous chapters though.
So we focus on the particular solution x
p
(t). We assume that it can be written as
x
p
(t) = x
p
cos t. (3.1.3)
Inserting this in the dierential equation will give
x
p
=

F
e
m
1
(
2
n

2
)
x
p
(t) =

F
e
m
1
(
2
n

2
)
cos t. (3.1.4)
If we combine this with the general solution to the homogeneous problem, we nd that
x(t) =
v
0

n
sin
n
t +x
0
cos
n
t +

F
e
m
1
(
2
n

2
)
(cos t cos
n
t) . (3.1.5)
A very important thing can be noticed from this equation. If
n
, then x
p
(t) and thus also
x(t) . This phenomenon is called resonance and is dened to occur if =
n
. It is something
engineers should denitely prevent.
3.2 Resonance
When looking at equation 3.1.5 we can see that it is undened for =
n
. What happens if we force a
system to vibrate at its natural frequency? To nd this out, we set =
n
. The dierential equation
now becomes
x +
2
n
x(t) =

F
e
m
cos
n
t. (3.2.1)
If we try a solution of the form x
p
(t) = x
p
cos
n
t, we will only nd the equation 0 =

F
e
/m

cos
n
t.
So there are no solutions of the assumed form. Instead, lets try to assume that x
p
(t) = x
p
t sin
n
t. We
now nd that
x
p
=

F
e
2m
n
x
p
(t) =

F
e
2m
n
t sin
n
t. (3.2.2)
What we get is a vibration in which the amplitude increases linearly with time. So as the time t
increases, also the amplitude of the motion increases. This continues until the system cant sustain the
large amplitudes anymore and will fail.
9
3.3 Beat Phenomenon
When the external force isnt vibrating at exactly the natural frequency of a system, but only close to it,
also interesting things occur. First lets dene the two variables and as
=

n

2
and =

n
+
2
. (3.3.1)
Lets once more consider equation 3.1.5. If we have no initial displacement or velocity (x
0
= 0 and
v
0
= 0), then we can rewrite this equation to
2

F
e
m
1
(
2
n

2
)
sin (t) sin ( t) = 2

F
e
m
1
(
2
n

2
)
sin

2
T
1
t

sin

2
T
2
t

. (3.3.2)
As
n
also 0 and
n
. So it follows that T
1
will become very large, while T
2
is close to
the natural frequency of the system. Since T
1
is so large, we can dene the amplitude of the vibration as
A(t) = 2

F
e
m
1
(
2
n

2
)
sin

2
T
1
t

. (3.3.3)
So we now have a rapid oscillation with a slowly varying amplitude. This phenomenon is called the beat
phenomenon and one variation of the amplitude is called a beat. As the forcing frequency goes closer
to the natural frequency
n
, both the amplitude and the period of a beat increase.
3.4 Harmonic Excitation of Damped Systems
Lets involve damping in our equations. We then get
m x +c x +kx =

F
e
cos t x + 2
n
x +
2
n
x =

F
e
m
cos t. (3.4.1)
Lets assume our particular solution can be written as
x
p
(t) = X cos (t ) . (3.4.2)
Inserting this in the dierential equation, and solving for X and , will eventually give
X =

F
e
m
1

(
2
n

2
)
2
+ (2
n
)
2
and = arctan

2
n

2
n

. (3.4.3)
To nd the general solution set, add x
p
(t) up to the solution of the homogeneous equation and use initial
conditions to solve for the coecients A and .
Lets dene the (dimensionless) frequency ratio as
r =

n
. (3.4.4)
We can now rewrite X and to
X =

F
e
k
1

(1 r
2
)
2
+ (2r)
2
and = arctan

2r
1 r
2

. (3.4.5)
If r 1 then X goes to a given maximum value. This maximum value strongly depends on the damping
ratio . For large values of , resonance is hardly a problem. However, if is small, resonance can still
occur.
10
3.5 Sinusoidal Forcing Functions
We have up to know only considered forcing functions involving a cosine. Of course forcing functions can
also be expressed using a sine. Lets examine the forcing function
F
e
(x) =

F
e
sin t. (3.5.1)
The particular solution to the (damped) dierential equation then becomes
x
p
(t) = X sin (t ) . (3.5.2)
The variables X and are still the same as in equation 3.4.5.
3.6 Base Excitation
Lets now suppose no external force is acting on the mass. Instead the base on which the spring is
connected, is moving by an amount x
b
(t), as shown in gure 3.5.
Figure 3.5: Denition of variables in base excitation.
The elongation of the spring is now not given by just x(t), but by x(t) x
b
(t). Identically, its velocity
with respect to the ground is now x(t) x
b
(t). So this makes the dierential equation describing the
problem
x + 2
n
x +
2
n
x = 2
n
x
b
(t) +
2
n
x
b
(t). (3.6.1)
Often the base excitation is harmonic, so we assume that
x
b
(t) = x
b
sin
b
t. (3.6.2)
This makes the dierential equation
x + 2
n
x +
2
n
x = 2
n

b
x
b
cos
b
t +
2
n
x
b
sin
b
t. (3.6.3)
We have two nonhomogeneous parts. We can therefore nd two separate particular solutions for the
dierential equation (one for each part). If we set

F
e
/m = 2
n

b
(or identically

F
e
/k = 2r), then we
have exactly the same problem as we have seen earlier with the cosine forcing function (equation 3.4.2).
If we, on the other hand, set

F
e
/m =
2
n
x
b
(or identically

F
e
/k = x
b
), then we have the same problem
as we just saw with the sine forcing function (equation 3.5.2). Add the two solutions up to get the total
particular solution
x
p
(t) =
2r x
b

(1 r
2
)
2
+ (2r)
2
cos (t ) +
x
b

(1 r
2
)
2
+ (2r)
2
sin (t ) . (3.6.4)
The value of is still the same as it was in equation 3.4.5.
11
4. General Forced Vibrations
4.1 The Impulse Function
An impulse excitation is a force that is applied for a very short duration t with respect to the vibration
period T = 2/
n
. It is an example of a shock loading. Such an impulse can be mathematically
represented by using the unit impulse function (t) (also called the Dirac delta function), dened
such that
(t ) = 0 for t = , (4.1.1)

(t )dt = 1. (4.1.2)
But how does this eect the motion of a system? Lets suppose we have a system with no initial
displacement and mass, that is given an impulse

F
e
at time t = . The corresponding dierential
equation is
m x +c x +kx = F
e
(t ). (4.1.3)
This impulse will cause the linear momentum of the mass to change by

F
e
= F
e
t = mv = mv

. (4.1.4)
So this situation is similar to the case where the object simply has an initial velocity of v

at time t =
(with x

= 0). If we apply this, for example, to an underdamped system, we would get the equation of
motion
x(t) =

F
e
h(t ), where h(t) =
1
m
d
e
nt
sin
d
t. (4.1.5)
The function h(t) is now called the impulse response function.
4.2 The Step Function
Another case of a forcing function is the unit step function u(t) (also called the Heaviside step
function, dened such that
u(t ) =

0 for t < ,
1 for t .
(4.2.1)
Lets consider the underdamped dierential equation
m x +c x +kx =

F
e
u(t ). (4.2.2)
If x
0
= 0 and v
0
= 0, it can be shown that
x(t) =

F
e
k

1
1

1
2
e
nt
cos (
d
t )

, (4.2.3)
where is given by
= arctan

1
2

. (4.2.4)
This solution looks awfully familiar. In fact, it corresponds to a vibration with equilibrium point x
e
=

F
e
/k and initial displacement x
0
= 0.
12
4.3 Replacing a Periodic Forcing Function by a Fourier Series
What if we dont have just an impulse or a step function, but a continuous forcing function F
e
(t)? In
this case we can take the force F
e
() at time for a given moment d and replace it by an impulse of
magnitude F
e
()d. We can then nd the impulse response function h(t ) for the time . If we do
this for all times and sum everything up, we will eventually nd as particular solution
x
p
(t) =

t
0
F
e
()h(t )d =

t
0
F
e
(t )h()d. (4.3.1)
This integral is called the convolution integral. It is often dicult to evaluate the integral. If we have
a periodic forcing function F
e
(t) (with period T and angular frequency
T
= 2/T), we can apply a trick
though. We can replace F
e
(t) by a Fourier series. To do this, we use
F
e
(t) =
a
0
2
+

n=1

a
n
cos

n
2
T
t

+b
n
sin

n
2
T
t

. (4.3.2)
The coecients a
0
, a
n
and b
n
are given by
a
0
=
2
T

T
0
F
e
(t)dt, (4.3.3)
a
n
=
2
T

T
0
F
e
(t) cos

n
2
T
t

dt, (4.3.4)
b
n
=
2
T

T
0
F
e
(T) sin

n
2
T
t

dt. (4.3.5)
Now we have a new way to write the forcing function. How we use this will be treated in the next
paragraph.
4.4 Finding the Equation of Motion
When we replace the periodic forcing function F
e
(t) by a Fourier Series, we can rewrite the dierential
equation to
m x +c x +kx =
a
0
2
+

n=1
(a
n
cos (n
T
t) +b
n
sin (n
T
t)) . (4.4.1)
We now repeatedly take one element from the right hand side of the equation, solve the equation for that
part, and in the end sum everything up. We will then nd our particular solution. In an equation this
becomes
x
p
(t) = x
a0
(t) +

n=1
(x
an
(t) +x
bn
(t)) . (4.4.2)
The individual solution are then the solutions of the dierential equations
m x
a0
+c x
a0
+kx
a0
= a
0
/2, (4.4.3)
m x
an
+c x
an
+kx
an
= a
n
cos (n
T
t) , (4.4.4)
m x
bn
+c x
bn
+kx
bn
= b
n
sin (n
T
t) . (4.4.5)
13
All these equations are equations we have solved before. For completeness sake we will give the solutions
once more. They are
x
a0
=
a
0
2k
, (4.4.6)
x
an
=
a
n
m
X cos (n
T
t
n
) , (4.4.7)
x
bn
=
b
n
m
X sin (n
T
t
n
) . (4.4.8)
The variables X and
n
are dened as
X =
1

2
n
(n
T
)
2

2
+ (2n
n

T
)
2
and
n
= arctan

2n
n

2
n
(n
T
)
2

. (4.4.9)
This is how the particular solution is found. Combine this with the specic solution to the problem to
nd the general solution to the dierential equation.
4.5 Using the Laplace Transform
When solving the dierential equation, the Laplace transform is often a convenient tool. Lets consider
the dierential equation
m x +c x +kx = F
e
(x) x + 2
n
x +
2
n
x =
F
e
(x)
m
. (4.5.1)
Taking the laplace transform, and solving for X(s), will give
X(s) =
sx
0
+v
0
+ 2
n
x
0
s
2
+ 2
n
s +
2
n
+
1
m

e
(s)
s
2
+ 2
n
s +
2
n
, (4.5.2)
where L{F
e
(t)} =
e
(s). Often it occurs that x
0
= 0 and v
0
= 0. The middle term of the above equation
then disappears. To nd x(t), you apply the inverse Laplace transform. When doing this, you often need
to use a Laplace transform table like table 4.1.
Function x(t) = L
1
{X(s)} Laplace Transform X(s) = L{x(t)} Condition
e
at 1
s+a
sin
n
t
a
s
2
+
2
n
cos
n
t
s
s
2
+
2
n
1
s
2
+2ns+
2
n
1

d
e
nt
sin (
d
t) Underdamped Motion ( < 1)

2
n
s
1
s
2
+2ns+
2
n
1
1

1
2
e
nt
sin (
d
t + arccos ()) Underdamped Motion ( < 1)
e
at
x(t) X(s +a)
(t a) e
as
u(t a)x(t) e
as
X(s)
Table 4.1: Often used Laplace transforms.
14
5. Multiple-Degree-of-Freedom Systems
5.1 Governing Equations of a Two-Degree-of-Freedom System
In previous chapters we have only looked at systems with one changing variable x. In reality situations
can hardly ever be expressed by just one variable. To investigate multiple-degree-of-freedom systems, we
will rst look at two-degree-of-freedom systems. An example of such a system is shown in gure 5.6.
Figure 5.6: An example of a two-degree-of-freedom system.
When drawing the equations of motion for each mass, the general equations of motion can be derived.
These are
m
1
x
1
= k
1
x
1
+k
2
(x
2
x
1
) , (5.1.1)
m
2
x
2
= k
2
(x
2
x
1
) . (5.1.2)
(We are not considering damping for multiple-degree-of-freedom systems.) When solving this system,
four boundary conditions are necessary. These are x
10
, x
10
, x
20
and x
20
.
However, writing things like this is a bit annoying. Its better to use vectors and matrices. First lets
dene the position vector x, the velocity vector x and the acceleration vector x as
x =

x
1
x
2

, x =

x
1
x
2

and x =

x
1
x
2

. (5.1.3)
We can also dene the mass matrix (also called the inertia matrix) for two-degree-of-freedom cases
as
M =

m
1
0
0 m
2

. (5.1.4)
Finally we also need the stiness matrix. For our example system, this matrix is
K =

k
1
+k
2
k
2
k
2
k
2

. (5.1.5)
Now we can write the system of dierential equations as
M x +Kx = 0. (5.1.6)
Note that both M and K are symmetric matrices (meaning that M
T
= M and K
T
= K). M is symmetric
because all non-diagonal terms are simply zero. K is symmetric due to Newtons third law.
We now want to nd the equation of motion x(t) for the system of dierential equations. To get it, we
need to solve equation 5.1.6. There are multiple ways to do this. Well discuss two ways.
15
5.2 First Method to nd the Equation of Motion
The rst method we will be discussing is usually the simplest method for hand calculation. It is therefore
quite suitable for applying on examinations. Computers, however, dont prefer this method.
Lets suppose our solution has the form x(t) = ue
it
. Filling this in into the dierential equation will
give

K
2
M

ue
it
= 0. (5.2.1)
The exponential cant be zero. Also, if u = 0, we wont have any motion either. So we need to have
such that the matrix

K
2
M

is singular (not invertible). In other words, its determinant must be

zero. The characteristic equation then is
det

K
2
M

= 0. (5.2.2)
For our two-degree-of-freedom example system, this results in
m
1
m
2

4
(m
1
k
2
+m
2
k
1
+m
2
k
2
)
2
+k
1
k
2
= 0. (5.2.3)
From this equation four values of will be found, being
1
and
2
. These are the natural frequen-
cies of the system. So although a one-degree-of-freedom has only one natural frequency, a two-degree-
of-freedom system has 2 natural frequencies. Multiple-degree-of-freedom systems have even more natural
frequencies.
The corresponding (nonzero) vectors u
1
and u
2
can now be found using

K M
2
1

u
1
= 0 and

K M
2
2

u
2
= 0. (5.2.4)
Only the direction of the vectors u can be derived from the above relations. Their magnitudes may be
chosen arbitrarily, although they are often normalized such that ||u|| = 1. The nal equation of motion
is then given by
x(t) = A
1
sin (
1
t +
1
) u
1
+A
2
sin (
2
t +
2
) u
2
. (5.2.5)
The values of A
1
,
1
, A
2
and
2
now need to be determined from the initial conditions.
5.3 Second Method to nd the Equation of Motion
There is another way to nd the equation of motion. Before we discuss this method, we rst have to
make some denitions. We dene the matrix square root M
1/2
of M such that
M
1/2
M
1/2
= M M
1/2
=

m
1
0
0

m
2

. (5.3.1)
This matrix also has an inverse

M
1/2

1
= M
1/2
. Lets dene the vector q such that
x(t) = M
1/2
q(t). (5.3.2)
Lets assume q = ve
it
, with v a constant vector. We can now rewrite equation 5.1.6 to
M
1/2
KM
1/2
v =

Kv =
2
v, (5.3.3)
where

K = M
1/2
KM
1/2
is the mass normalized stiness. If we replace
2
by in the above
equation we have exactly the eigenvalue problem from linear algebra. The solutions for are then the
eigenvalues of the matrix

K and the corresponding vectors v are the eigenvectors.
16
Since K is symmetric, also

K is symmetric. All the eigenvalues are therefore real numbers and also the
eigenvectors are real. Once the eigenvalues
1
and
2
are known, the natural frequencies
1
and
2
can
easily be found using

1
=

1
and
2
=

2
. (5.3.4)
To nd the corresponding vectors u, you can use
u
1
= M
1/2
v
1
and u
2
= M
1/2
v
2
(5.3.5)
The equation of motion is then once more given by
x(t) = A
1
sin (
1
t +
1
) u
1
+A
2
sin (
2
t +
2
) u
2
. (5.3.6)
5.4 Modal Analysis
We can also nd the equation of motion using modal analysis. In the previous paragraph we have found
the eigenvectors v
1
and v
2
of the matrix

K. These vectors are orthogonal (unless they correspond to
the same eigenvalue, in which case they should be made orthogonal). If they have also been normalized
(given length 1), then they form an orthonormal set. Now lets dene the matrix of eigenvectors P
to consist of these orthonormal eigenvectors. In an equation this is
P =

v
1
v
2

. (5.4.1)
This matrix is an orthogonal matrix (as its columns are orthonormal). Such matrices have the conve-
nient property that P
T
P = I. Also lets dene the matrix of mode shapes S as
S = M
1/2
P. (5.4.2)
Furthermore we dene the vector r(t) such that
x(t) = M
1/2
q(t) = M
1/2
Pr(t) = Sr(t). (5.4.3)
Using all these denitions, we can rewrite the system of dierential equations to
r(t) + r(t) = 0, (5.4.4)
where the matrix is given by
= P
T

KP =

2
1
0
0
2
2

. (5.4.5)
So we remain with the dierential equations
r
1
+
2
1
r = 0, (5.4.6)
r
2
+
2
2
r = 0. (5.4.7)
The dierential equations have been decoupled! They dont depend on each other, and therefore can be
solved using simple methods. The two decoupled equations above are called the modal equations. Also
the coordinate system r(t) is called the modal coordinate system.
To solve the modal equations, we need the initial conditions in the modal coordinate system. Usually we
only know the initial conditions x
0
and x
0
in the normal coordinate system. We can transform these to
the modal coordinate system using
r
0
= S
1
x
0
and r
0
= S
1
x
0
, where S
1
= P
T
M
1/2
. (5.4.8)
Now we can solve for r
1
(t) and r
2
(t) and thus for r(t). Once we have found r(t) we can nd the equation
of motion x(t) using
x(t) = Sr(t). (5.4.9)
17